Pinecone query takes forever

Support
1

We’re using a Pinecone Index where each vector contains a ‘content’ metadata field with around 6,800 characters. When running the following query:

const similarEmbeddings = await PI.query({
      vector,
      topK: 12,
      includeMetadata: true,
});

The query seems to hang indefinitely. However, when we reduce topK to less than 5, the results return as expected.

Are there any limitation to amount of data we can store in metadata and can fetch while running a query?

2

Hi @tech, and welcome to the Pinecone community forums!

Thanks for your question.

  1. Where is your code running?
  2. Where is your index hosted (region)
  3. Which kind of index is it? (Serverless / pods)
  4. We do have limits, which are published here, but I would not immediately expect your query to be hitting a limit based off what you’ve shared so far.

Any other relevant code you can share would also be helpful.

Best,
Zack

3

@ZacharyProser

  1. Where is your code running? - Env is Node.js. We tested in local and EC2 (Mumbai region)
  2. Where is your index hosted (region) - us-east-1
  3. Which kind of index is it? (Serverless / pods) - Serverless
  4. We do have limits, which are published, but I would not immediately expect your query to be hitting a limit based off what you’ve shared so far - Certainly our Metadata isn’t even near 4KB limit

Additional context - Our Index has 34 records, each record has - embedding and metadata - metadata has 3 fields, one of them is content. Content has data used to generate embedding in the first place and on average is 7000 characters in length.

Additional debugging

  • When we run query without includeMetadata: true, the query is fast and works as expected. Seems like there is an issue with Metadata.